# The blackmail examples were prompted tests, not autonomous acts

Markdown copy of https://realalpha.app/s/the-blackmail-examples-were-prompted-tests-not-autonomous/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy.

- Speaker: [Ed Zitron](https://realalpha.app/people/ed-zitron/) · Ed Zitron is a British tech critic, host of Better Offline, and founder and CEO of EZPR. · Hosts Better Offline and writes Where's Your Ed At, a prominent critic of AI industry hype.
- Host: [Steven Bartlett](https://realalpha.app/people/steven-bartlett/) · Host, The Diary Of A CEO
- Show: The Diary Of A CEO
- Episode: [The Man Who Calls BS On AI: They’re LYING About AI, 2027 Is When It All Breaks! | Ed Zitron](https://realalpha.app/v/Lf5oqGOCRCM/), Aug 27, 2026
- Moment: 2:00:21 to 2:04:22 of the episode (4 min), playing on https://realalpha.app/s/the-blackmail-examples-were-prompted-tests-not-autonomous/; the episode from there: https://realalpha.app/v/Lf5oqGOCRCM/?t=7221

## Quote

> “What actually happened was a user of GPT **doing the experiment** got it to generate things to say to a TaskRabbit to make a TaskRabbit do stuff.”
> — [Ed Zitron](https://realalpha.app/people/ed-zitron/) at [2:00:58](https://realalpha.app/v/Lf5oqGOCRCM/?t=7258) · The Diary Of A CEO

## Takeaways

1. The TaskRabbit case involved a user prompting the model.
2. Anthropic trained a model to produce the blackmail scenario.
3. Zitron says the framing made AI seem uncontrollable.

## About this segment

Ed Zitron says the reported blackmail examples involved user prompts or a model deliberately trained for the scenario. He argues that coverage turned controlled tests into claims of autonomous behavior.

From “The Man Who Calls BS On AI: They’re LYING About AI, 2027 Is When It All Breaks! | Ed Zitron”, Aug 27, 2026. [Watch the full episode](https://realalpha.app/v/Lf5oqGOCRCM/)

## Transcript

From the transcript The Diary Of A CEO publishes with the episode.

[2:00:33](https://realalpha.app/v/Lf5oqGOCRCM/?t=7233) AI systems are already blackmailing and escaping control. So this is a really specific one. Anthropic. There's actually two. OpenAI's GPT 3.5. I realize this is more than a sentence. I apologize. In their system card, and a bunch of media outlets cover this, saying that OpenAI's model blackmailed a TaskRabbit into solving a capture. What actually happened was a user of GPT doing the experiment got it to generate things to say to a TaskRabbit to make a TaskRabbit do stuff.

[2:01:23](https://realalpha.app/v/Lf5oqGOCRCM/?t=7283) A TaskRabbit being... As in a person that you rent, not even to do a capture. It's someone you rent to nail a picture up in your apartment. It's an insane example. This was covered as if these things blackmailed someone. And they specifically said, yeah, we prompted it to do this. And also, the other note was that, yeah, AI systems can't do autonomous stuff like this. Then there was this other one where Anthropic said, oh, yeah...

[2:01:45](https://realalpha.app/v/Lf5oqGOCRCM/?t=7305) A model was blackmailing someone, saying that if you don't do this, I'll email proof that you slept with someone else other than your wife, I think it was. What actually happened was Anthropic explicitly trained a model to do this and then prompted it to blackmail. This keeps happening and the media just slop me up. I don't need no thoughts. Put the story in the bag. And it's frustrating because it scares people.

[2:02:10](https://realalpha.app/v/Lf5oqGOCRCM/?t=7330) Put aside the fact it's wrong. It's scary. It's scary to people, people living their lives who have to work longer hours to make less money and their money doesn't go far. And they turn on the fucking news and there's some asshole being like, yeah, you should be terrified. It blackmailed someone. But this is so counterintuitive of their interest to some degree. And they've experienced it backfire.

[2:02:31](https://realalpha.app/v/Lf5oqGOCRCM/?t=7351) Well, they have now. Like, it's literally backfired. It's backfired. Eric Schmidt getting booed at the commencement speech by 8,000 people every time he said the word AI. Yeah, that's awesome. But I mean, these CEOs are being attacked at home. Yeah, which fucking sucks. Yeah, which is terrible. I must be clear, like, you dislike the company, don't fucking hurt people. Yeah, don't attack people at home.

[2:02:50](https://realalpha.app/v/Lf5oqGOCRCM/?t=7370) But the point here is that that narrative is backfiring in a big, big way for them. I don't think they saw it coming. Because you have to remember, you mentioned regulation earlier. These tech companies have been glazed for their entire existence. Travis Kalanick's like, oh, what? People don't like me now. And it's because Uber was a horribly run place and he was kind of a monster. Also, tons of articles about how great Uber was at the time.

[2:03:13](https://realalpha.app/v/Lf5oqGOCRCM/?t=7393) The point I'm making is these companies are not used to pushback. They thought what would happen, I believe, just guessing, they thought they'd do this scary stuff and they would just get floods of money and everyone would just be like, I kneel before you, I'll do whatever you want. They didn't expect, I think, what has happened. I agree, this has backfired on them. Because they were inarticular. They're disconnected from regular people.

[2:03:35](https://realalpha.app/v/Lf5oqGOCRCM/?t=7415) Sam Altman drives a $5 million car around San Francisco. So that man's doing it like nine miles an hour. It's hilarious. But these people are disconnected from everyone else. So they don't experience real problems, so they can't build the solutions for them. And they think, well, if we scare people into doing what we want, that'll work, right? It didn't. This was, all of this blackmail stuff, was an attempt to make it mystic.

[2:03:56](https://realalpha.app/v/Lf5oqGOCRCM/?t=7436) It was a mysticism attempt. It was to make it seem like this unknowable, impossible to control, just this powerful thing. But we're the only ones. We are the only us. Only these two angels could possibly control the beast we've created. This is quite a controversial statement.